Re: md5sum and binary vs. text

[Jim Meyering]

Eric Blake wrote:
> According to Jim Meyering on 3/25/2009 3:57 PM:
>>> and the extra * breaks the check.  I'd really like to switch md5sum and
>>> friends to match Linux output when O_BINARY, and add a different marker
>>> character (but what?) when summing a file in O_TEXT mode, since generally
>>> you WANT binary mode to be the default.  Thoughts?
>>
>> I see what you mean, but what about people who use -b on Linux.
>> They'd still have to get the "*".
>
> People who use -b on Linux would see no change in behavior, because there,
> like any other sane system, text and binary mode files are
> indistinguishable, so there is no need for a marker.

Right, the *checksum values* would not change, but currently,
these tools always print the "*" marker with an explicit -b:

  $ :|sha1sum -b
  da39a3ee5e6b4b0d3255bfef95601890afd80709 *-

so I'm afraid we'll have to retain that behavior.

>> Seems like it'd be hard to arrange such a change
>> without causing too much disruption.
>
> It seems to me that only platforms like Cygwin would be impacted.  In
> fact, one of the commits that impacts binary mode (4d87cd7, back in 2001)
> is attributed to Chris Faylor, who was the person responsible for porting
> coreutils to cygwin back in the days where the cygwin community was not
> very sure about text vs. binary decisions; since then, the cygwin
> community has stabilized to preferring binary over text as a default.
>
> For md5sum -c purposes, we will have to continue accepting * in files that
> we validate.  But for generation purposes, it seems like we should
> generate space (binary, and make it the default on cygwin) or a new marker
> (text, when requested, and only on platforms where it matters).  And since
> the BSD format is different altogether, it seems like only GNU md5sum
> would be impacted by a change in the marker character.